CN109509170A - A kind of die casting defect inspection method and device - Google Patents
A kind of die casting defect inspection method and device Download PDFInfo
- Publication number
- CN109509170A CN109509170A CN201811056776.XA CN201811056776A CN109509170A CN 109509170 A CN109509170 A CN 109509170A CN 201811056776 A CN201811056776 A CN 201811056776A CN 109509170 A CN109509170 A CN 109509170A
- Authority
- CN
- China
- Prior art keywords
- die casting
- defect
- image
- characteristic
- algorithm
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 230000007547 defect Effects 0.000 title claims abstract description 200
- 238000004512 die casting Methods 0.000 title claims abstract description 148
- 238000000034 method Methods 0.000 title claims abstract description 67
- 238000007689 inspection Methods 0.000 title claims abstract description 29
- 238000001514 detection method Methods 0.000 claims abstract description 30
- 238000012549 training Methods 0.000 claims abstract description 21
- 230000002708 enhancing effect Effects 0.000 claims abstract description 10
- 238000012545 processing Methods 0.000 claims abstract description 6
- 230000008569 process Effects 0.000 claims description 25
- 230000003044 adaptive effect Effects 0.000 claims description 16
- 238000000354 decomposition reaction Methods 0.000 claims description 12
- 238000005516 engineering process Methods 0.000 claims description 11
- 239000000284 extract Substances 0.000 claims description 10
- 238000004519 manufacturing process Methods 0.000 claims description 8
- 238000005070 sampling Methods 0.000 claims description 8
- 230000009466 transformation Effects 0.000 claims description 8
- 238000013527 convolutional neural network Methods 0.000 claims description 6
- 238000004458 analytical method Methods 0.000 claims description 5
- 230000004913 activation Effects 0.000 claims description 4
- 238000005266 casting Methods 0.000 claims description 3
- 238000003860 storage Methods 0.000 claims description 3
- 238000013519 translation Methods 0.000 claims description 2
- 238000006073 displacement reaction Methods 0.000 claims 1
- 238000001914 filtration Methods 0.000 claims 1
- 238000011002 quantification Methods 0.000 claims 1
- 230000002950 deficient Effects 0.000 abstract description 4
- 230000004044 response Effects 0.000 description 8
- 230000008859 change Effects 0.000 description 5
- 230000006870 function Effects 0.000 description 4
- 238000011161 development Methods 0.000 description 3
- 230000007306 turnover Effects 0.000 description 3
- 230000015572 biosynthetic process Effects 0.000 description 2
- 230000001413 cellular effect Effects 0.000 description 2
- 238000010276 construction Methods 0.000 description 2
- 125000004122 cyclic group Chemical group 0.000 description 2
- 238000009826 distribution Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000007613 environmental effect Effects 0.000 description 2
- 238000005286 illumination Methods 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 230000004660 morphological change Effects 0.000 description 2
- 238000012797 qualification Methods 0.000 description 2
- 238000013139 quantization Methods 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 230000008901 benefit Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000005352 clarification Methods 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 238000012417 linear regression Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 210000005036 nerve Anatomy 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 230000002093 peripheral effect Effects 0.000 description 1
- 238000012805 post-processing Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
- G06T7/0008—Industrial image inspection checking presence/absence
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/70—Denoising; Smoothing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20172—Image enhancement details
- G06T2207/20192—Edge enhancement; Edge preservation
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of die casting defect inspection method and devices, there is the die casting picture of defect characteristic by acquiring, defect picture is denoised, enhancing processing, with defect picture training depth convolutional network Alexnet, classified to defect picture, establish corresponding semantic label and is stored to defect characteristic library, Bug Tracking is carried out to die casting in real time, when detecting the defective die casting of tool, feeds back to detection device and statistic mixed-state data, generation die casting defects detection report.The present invention improves die casting defects detection efficiency and precision, and accuracy rate is high, and speed is fast, and real-time is good, can fast and accurately find defective die casting.
Description
Technical field
The present invention relates to technical field of quality detection, a kind of die casting defect inspection method and device are referred in particular to, it is readable
Storage medium and computer control system.
Background technique
Quality of die casting detects the important component as die casting post-processing stages, and detection efficiency and quality are direct
It is related to the quality of final products.Traditional quality of die casting detection method is detected to die casting surface defect, generally
For artificial detection, artificial detection is the detection method of static part, can only specific point and position be detected and be measured,
In order to improve measurement accuracy and accuracy, generally requires multiple workers and repeatedly measure averaged.In detection process completely
It goes to carry casting by worker oneself, be classified according to the result that worker measures to product.And artificial detection detection effect
Rate is low, detects bad environments, is easy to be influenced by worker's subjective factor.With the development of industrial technology, based on machine vision from
Dynamicization defect detection equipment comes into being, and still, is currently based on defects of vision detection method also in initial development stage, detection
Technology is also immature, and defect inspection method system is also imperfect.Especially in die casting defects detection industry, due to die casting table
The defect characteristic in face is very strange, in fact it could happen that the different defect characteristic such as greasy dirt, scratch, overlap, burr, in detection process
In, different defect characteristics may interfere with each other, and cause accidentally to survey, for example the defect characteristic for needing to detect is obvious hair
Die casting with greasy dirt feature may be also included into together testing result by thorn.
Summary of the invention
For the problems in above-mentioned background technology, a kind of die casting defect inspection method is provided, quickly can accurately be examined
The surface defect of high-volume die casting is measured, to assess the quality of production of die casting.
A kind of die casting defect inspection method of the present invention, comprising:
S1 obtains the defect image data of die casting in actual production；
S2 extracts the defects of defect image feature；
S3 classifies to the defect characteristic；
S4 establishes corresponding semantic label to the classification of the various defect characteristics；
The defect characteristic and corresponding semantic label are stored in property data base by S5；
S6 shoots the video image on die casting surface, and is converted into image data；
S7 closes filter tracks algorithm using nuclear phase and tracks to the image data, in the property data base
Defect characteristic is drawbacks of the standard feature, records the drawbacks of the standard feature having in the image data traced into；
S8 counts ratio shared by the drawbacks of the standard characteristic and different drawbacks of the standard features of the image data.
Defect image: the die casting surface defect video or picture obtained for industrial camera shooting.
Defect characteristic: for the defect characteristic on the die casting surface identified in defect image.
Semantic label: the structure and features facilitated the search for defines label.
Property data base: the database to store die casting defect image and defect characteristic semantic label.
The present invention, which passes through acquisition die casting defect characteristic and classifies, is stored in database, and nuclear phase is used in defect inspection process
The method for closing the defects of the database that filter tracks algorithm record die casting has feature, can rapidly and accurately find
The die casting of defect improves die casting defects detection efficiency and precision, and accuracy rate is high, and speed is fast, and real-time is good.
It specifically, include: input needs by the step of drawbacks of the standard feature of the defects of property data base feature
The die casting defect title of detection, searches for matching defect characteristic, the defect characteristic that will be searched in property data base
As the drawbacks of the standard feature.
The defects of property data base feature has been subjected to classification and establishes semantic label, to need to examine in defect inspection process
The die casting defect name of survey is referred to as search key, and to be detected using the defect characteristic searched as drawbacks of the standard feature
Die casting defect characteristic compare and analyze, detection efficiency can be improved, targetedly a certain defect characteristic is detected.
Further, before classifying to the defect characteristic, at Wavelet Algorithm and image enhancement technique
Manage the defects of defect image feature；Wherein, the Wavelet Algorithm includes with Nonlinear Wavelet Transform threshold method to die casting
The defect image of part carries out denoising, carries out multiscale analysis to die casting image with the operation of flexible and shift operations,
The automatic requirement adapted in different type die casting and light environment, shows the local feature of die casting；Described image enhancing
Technology includes enhancing algorithm with logarithmic image, distinguishes the defect and background of die casting image.
Die casting surface image often shows as Analysis On Multi-scale Features, after wavelet transformation, can obtain different resolution
Image, the image of different resolution is by representing the frequency subband image construction of different directions information.
Further, the step for carrying out denoising to the defect image of die casting with Nonlinear Wavelet Transform threshold method
Suddenly include: that wavelet basis and the wavelet decomposition number of plies is selected to carry out wavelet transformation to the defect image of die casting, obtain corresponding small echo
Decomposition coefficient, for each wavelet decomposition layer, adaptive one threshold value of selection will decompose obtained high frequency coefficient and carry out threshold value
Quantization, according to after wavelet decomposition low frequency coefficient and threshold value quantizing treated high frequency coefficient carry out wavelet inverse transformation, recycle
Restructing algorithm carries out wavelet reconstruction, the signal after being denoised.
Further, classify to the defect characteristic, corresponding language is established to the classification of the various defect characteristics
The step of adopted label includes: to input the defect image in Alexnet convolutional neural networks to be trained, using backpropagation
Algorithm acquires the minimum value of loss function f (χ, b), and wherein χ is weighted value, and b is biasing, using sigmoid functionAs activation primitive, weighted value is updated by more gradient Descent iteration methods and acquires optimal weights value, root
Corresponding weighted value { χ is inputted according to different defect characteristics1,χ2,…χn, shallow-layer extracts the preliminary feature of die casting defect, deep
Layer extracts the minutia of die casting defect, after training convolutional network model, classifies to die casting defect characteristic, in depth convolution
The end of network marks all kinds of die casting defect characteristics with softmax classifier, establishes corresponding semantic label.
Die casting defect characteristic is classified and establishes semantic label, is corresponded to convenient for being searched in subsequent defective detection process
Defect characteristic, and statistic of classification is carried out to the defect characteristic detected, with output test result.
Further, described to close the step of filter tracks algorithm tracks image data packet using nuclear phase
It includes: by carrying out cyclic shift to die casting Bug Tracking target area, constructing cluster sampling training classifier, calculate candidate regions
The similarity degree in domain and tracking target, the selection maximum candidate region of similarity are new tracking target, utilize discrete fourier
Transformation reduces the operand in classifier training and detection process.
Edge information formation can be changed by closing filter tracks algorithm using nuclear phase, become more fully apparent it, reached
To the purpose of enhancing die casting defect characteristic edge feature, die casting can be better described by the image that edge configuration is reinforced
The local detail of part such as can eliminate external environment such as light source, block at the influence to die casting image.
Further, it is to close filter tracks algorithm based on adaptive nuclear phase that the nuclear phase, which closes filter tracks algorithm,
Its step includes: to adjust the weighted value of present frame by being updated to classifier, and target is carried out piecemeal, calculates each point
The PSR value of block, weight of each localized mass of adaptively changing to target position.
During tracking to die casting, the surface of die casting may be with such as illumination of certain environmental factors
Change and change, this is there is a possibility that tracking accuracy reduces, and adaptive nuclear phase closes the avoidable appearance of filter tracks algorithm
Such case improves tracking effect, avoids external interference.
The present invention also provides a kind of die casting defect detecting devices, comprising:
For obtaining the device of the defect image data of die casting in actual production；
For extracting the device of the defects of defect image feature；
Device for classifying to the defect characteristic；
The device of corresponding semantic label is established for the classification to the various defect characteristics；
For the defect characteristic and corresponding semantic label to be stored in the device of property data base；
For shooting the device of the video image on die casting surface and being translated into the device of image data；
Device for being tracked by nuclear phase pass filter tracks algorithm to the image data, and record trace into
The image data in the device of drawbacks of the standard feature that has；
For counting ratio shared by the drawbacks of the standard characteristic and different drawbacks of the standard features of the image data
Device.
Further, the present invention also provides a kind of readable storage medium, control program is stored thereon, it is characterised in that:
The die casting defect inspection method as described in above-mentioned any one is realized when the control program is executed by processor.
Further, the present invention also provides a kind of computer control system, including reservoir, processor and it is stored in institute
State the control program that can be executed in reservoir and by the processor, it is characterised in that: the processor executes the control journey
The die casting defect inspection method as described in above-mentioned any one is realized when sequence.
In order to the clearer understanding present invention, a specific embodiment of the invention is illustrated below with reference to Detailed description of the invention.
Detailed description of the invention
Fig. 1 is a kind of die casting defect inspection method flow chart of the embodiment of the present invention.
Fig. 2 is to close filter tracks algorithm flow chart based on adaptive nuclear phase.
Specific embodiment
Referring to Fig. 1, it is a kind of die casting defect inspection method flow chart of the embodiment of the present invention.
A kind of die casting defect inspection method of the present invention, comprising:
S1 obtains the defect image data of die casting in actual production；
S2 extracts the defects of defect image feature；
S3 classifies to the defect characteristic；
S4 establishes corresponding semantic label to the classification of the various defect characteristics；The defect of fuzzy class is indicated with M,
Greasy dirt and it is water stain belong to fuzzy class defect, can be indicated with MY and MS, other kinds of defect and so on；
The defect characteristic and corresponding semantic label are stored in property data base by S5；
S6 shoots the video image on die casting surface, and is converted into image data；
S7 closes filter tracks algorithm using nuclear phase and tracks to the image data, in the property data base
Defect characteristic is drawbacks of the standard feature, records the drawbacks of the standard feature having in the image data traced into；
S8 counts ratio shared by the drawbacks of the standard characteristic and different drawbacks of the standard features of the image data.
The present invention, which passes through acquisition die casting defect characteristic and classifies, is stored in database, and nuclear phase is used in defect inspection process
The method for closing the defects of the database that filter tracks algorithm record die casting has feature, can rapidly and accurately find
The die casting of defect improves die casting defects detection efficiency and precision, and accuracy rate is high, and speed is fast, and real-time is good.
It specifically, include: input needs by the step of drawbacks of the standard feature of the defects of property data base feature
The die casting defect title of detection, searches for matching defect characteristic, the defect characteristic that will be searched in property data base
As the drawbacks of the standard feature.
Further, before classifying to the defect characteristic, at Wavelet Algorithm and image enhancement technique
Manage the defects of defect image feature；Wherein, the Wavelet Algorithm includes with Nonlinear Wavelet Transform threshold method to die casting
The defect image of part carries out denoising, carries out multiscale analysis to die casting image with the operation of flexible and shift operations,
The automatic requirement adapted in different type die casting and light environment, shows the local feature of die casting；Described image enhancing
Technology includes enhancing algorithm with logarithmic image, distinguishes the defect and background of die casting image.
Further, the step for carrying out denoising to the defect image of die casting with Nonlinear Wavelet Transform threshold method
Suddenly include: that wavelet basis and the wavelet decomposition number of plies is selected to carry out wavelet transformation to the defect image of die casting, obtain corresponding small echo
Decomposition coefficient, for each wavelet decomposition layer, adaptive one threshold value of selection will decompose obtained high frequency coefficient and carry out threshold value
Quantization, according to after wavelet decomposition low frequency coefficient and threshold value quantizing treated high frequency coefficient carry out wavelet inverse transformation, recycle
Restructing algorithm carries out wavelet reconstruction, the signal after being denoised.
Further, classify to the defect characteristic, corresponding language is established to the classification of the various defect characteristics
The step of adopted label includes: to input the defect image in Alexnet convolutional neural networks to be trained, using backpropagation
Algorithm acquires the minimum value of loss function f (χ, b), and wherein χ is weighted value, and b is biasing, using sigmoid functionAs activation primitive, weighted value is updated by more gradient Descent iteration methods and acquires optimal weights value, root
Corresponding weighted value { χ is inputted according to different defect characteristics1,χ2,…χn, shallow-layer extracts the preliminary feature of die casting defect, deep
Layer extracts the minutia of die casting defect, and the number of plies of convolutional neural networks is arranged according to specific defect, such as scratch, hair
The obvious defects of this kind of defect characteristic such as thorn, can use the convolutional neural networks of less level, and this kind of for color difference, greasy dirt
More multi-level convolutional neural networks can be used in the unconspicuous defect of defect characteristic.After training convolutional network model, to die casting
Defect characteristic classification, in the end of depth convolutional network, softmax classifier marks all kinds of die casting defect characteristics, foundation pair
The semantic label answered.
Will treated that die casting defect picture is input to depth convolutional network classifies to defect characteristic, the present invention adopts
Use the model for being similar to AlexNet neural network model as training die casting defect characteristic.AlexNet is by 5 Ge Juan bases, three
A full articulamentum and a softmax classifier composition.Using unsaturation nonlinear function ReLU function as activation primitive,
It is 0 by linearly correcting the certain data of pressure, so that the network model has sparsity, to can quickly receive in the training process
It holds back.It is asked in partial nerve network layer using local acknowledgement's normalization operations such as random erasure and overlapping pool to solve over-fitting
Topic carries out pondization operation after carrying out convolution operation to die casting picture again, so that die casting image has after translation, rotation
Better stability, then avoids over-fitting by way of overlapping pool.This convolutional network uses two when running training
GPU parallel training strategy, can accelerate training speed.
In the training process to parameter, study momentum turnover rate is set as 0.9, and parameter weight attenuation rate is set as
0.0005, then according to update rule formula:wi+1=wi+vi+1It is iterated training,
Wherein i indicates that the number of iteration, v are to update momentum, and ε is learning rate,It is trained DiTarget letter when (i-th of batch)
The directional derivative about w is counted in wiValue.The zero-mean gaussian that the weight (weights) of every layer network is all 0.01 with standard deviation
Distribution is to initialize, and the biasing (biases) of 2,4,5 convolutional layers, and the biasing of full articulamentum, is all initialized with 1,
His biasing is initialized with 0, and such initialization mode accelerates training process.Softmax classifier can be simultaneously to difference
Die casting defect classify.
Die casting defect picture is input to trained Alexnet network, passes through softmax points in the end of network
Class device classifies to die casting defect characteristic, and the defect characteristic classified finally is sticked corresponding semantic label, and deposit is special
Levy database.
Further, the nuclear phase closes filter tracks algorithm (KCF) are as follows: by die casting Bug Tracking target area
Cyclic shift is carried out, cluster sampling training classifier is constructed, calculate candidate region and tracks the similarity degree of target, is chosen similar
Spending maximum candidate region is new tracking target, is reduced in classifier training and detection process using discrete Fourier transform
Operand.It in t frame, is sampled near the Pt of current location, one recurrence device of training.The recurrence device can calculate a wicket
The response of sampling；It in t+1 frame, is sampled near previous frame position Pt, the response of each sampling is judged with the recurrence device, it will
Strongest sampling is responded as this frame position Pt+1.Tracking speed is improved using linear regression training in tracing process.It utilizes
Gradient orientation histogram feature (HOG) is used as die casting defect local direction gradient, location of pixels (x, y) in die casting image
The horizontal gradient with vertical direction is defined as:
In defect inspection process, to avoid external condition from normalizing the influence of die casting picture to gradient intensity
Processing.KCF tracker uses histogram of gradients feature to indicate as clarification of objective, the direction Density Distribution at HOG feature edge
Or histogram of gradients indicates, target gray figure is divided into multi cell group, calculates the histograms of oriented gradients in each cellular, it will
Each cellular carries out the description that statistics generates target image.Since die casting may be illuminated by the light or block in defect inspection process
It influences, HOG feature is more sensitive to die casting local edge region.The coefficient of KCF tracker fixation linearly updates classifier
Coefficient, the change that undated parameter can not be adaptive.
To avoid because of the problem for tracking inaccuracy caused by the environmental factors such as illumination in defect inspection process, the present embodiment is mentioned
For a kind of automatic adjusument scheme.
It is to close filter tracks algorithm, step packet based on adaptive nuclear phase that the nuclear phase, which closes filter tracks algorithm,
It includes: adjusting the weighted value of present frame by being updated to classifier, target is subjected to piecemeal, calculates the PSR of each piecemeal
Value, weight of each localized mass of adaptively changing to target position.
Canny marginal information is used to die casting image, edge information formation is changed, it is become more fully apparent,
Achieve the purpose that enhance die casting defect characteristic edge feature, pressure can be better described by the image that edge configuration is reinforced
The local detail of casting such as can eliminate external environment such as light source, block at the influence to die casting image.PSR is defined asWherein uiAnd δiThe respectively mean value and standard deviation of peak response position peripheral region response, β are to adjust
Coefficient is saved, is arranged according to the actual situation.Use above-mentioned PSR value as the criterion for judging whether to update, it can be in adaptive change
Update times.
Referring to Fig. 2, it is to close filter tracks algorithm flow chart based on adaptive nuclear phase.
Step 1: die casting defect characteristic title to be detected, pattern process computer are inputted from pattern process computer
It is scanned for from defect characteristic database, searches out corresponding defect characteristic, and return to pattern process computer.
Step 2: industrial camera shoots die casting video surface image from workbench, and by video image real-time Transmission
To pattern process computer, each parameter is initialized, converts picture for video data.
Step 3: filter tracks algorithm is closed using nuclear phase, die casting defect is tracked.
Step 3.1: die casting defect characteristic to be detected is inputted into multiple features core correlation tracker, and judges present frame
Whether it is first frame, if it is, going to Step 3.2, otherwise goes to Step 3.3.
Step 3.2: tracking die casting video file passes through Gauss nuclear mapping and circular matrix intensive sampling, quickly training
Regularization ridge regression classifier, acquires weight coefficient
Step 3.3: calculating response, and update position, and response utilizes formulaIt calculates,
Middle ⊙ indicates vector or the corresponding multiplication of matrix element, Indicate apparent.
Step 3.4: setting adjustment factor is β (parameter can be arranged according to the actual situation), calculates turnover rate parameter PSR,
Then according to the update classifier of turnover rate parameter adaptive.
Step 3.5: judging whether video data tracks and finish, finish if having tracked, and the defect target tracked is inputted
Information processing system continues iteration update if not terminating.
Step 4: according to ratio shared by the die casting fault detection data statistics qualification rate being collected into and all kinds of defects
Deng generating and export quality of die casting examining report.
A kind of die casting defect inspection method in the present embodiment, the first industrial camera by being located on conveyer belt
Obtain die casting surface to be detected real time video data, be transferred to pattern process computer, image computer to video data into
Row processing makes it be converted into picture, then samples to picture according to certain sample frequency, and the picture after sampling is carried out
Denoising of Nonlinear Wavelet Transform Threshold Value Method and with defect characteristic enhancing algorithm image enhancement is carried out to the defect picture of die casting
Processing.
In the actual production process, when carrying out the detection of certain defect to a certain die casting, from defect characteristic database
Relevant defect characteristic data are obtained as drawbacks of the standard feature.Industrial camera to pattern process computer conveying video data it
Before, it also needs to obtain real time picture frame data by above-mentioned Wavelet Algorithm and image enhancement technique processing operation.To processed
Picture frame use the process for closing filter tracks algorithm tracking target defect based on adaptive nuclear phase as follows:
The histogram of gradients feature of processed die casting image is obtained, the Canny edge feature of die casting image is extracted,
Morphological change is carried out to edge feature, the edge feature after morphological change is superimposed with original image, reaches edge enhancing
Effect.Die casting defect characteristic title to be detected is inputted, and judges whether present image is first frame, if it is, utilizing
Die casting defect characteristic to be detected and target position are trained regularization least square classifier, calculate response, more
New images frame number, passes through formulaUpdate times parameter, then adaptive update classifier are obtained, if
Present image is not first frame, then directly calculates response and update position.Finally judge whether die casting picture reads to finish, if
It has read and has finished, then terminator, otherwise continued iteration and update.
When a certain frame die casting picture have a certain type defect, then record defective feature die casting quantity and
Defect characteristic type counts die casting qualification rate after all die castings all detect.
Target following is the research direction in computer vision field very forward position.At computer technology and information
The development of reason technology, target following using more and more extensive, be related to the various fields such as military affairs, national defence, industry, medical treatment, agricultural,
Due to its broad application prospect and potential economic value, target following has become an important technology.The present invention mentions
A kind of die casting defect inspection method out, quickly can accurately detect the surface defect of high-volume die casting, to comment
Estimate the quality of production of die casting.
Compared with the existing technology, construction depth convolutional network of the present invention extracts die casting surface defects characteristic, and common
Detection die casting defect when the mode manually demarcated that uses compare, such method is more accurate efficiently, establishes die casting table
Planar defect property data base is conducive to collect the analysis of die casting last handling process data, provides data for die casting process modification
It supports, secondly, providing convenience for defect target following.On the other hand, filter tracks algorithm is closed compared to traditional nuclear phase
(KCF) method, adaptive classifier, which updates nuclear phase pass filter tracks algorithm, to keep KCF tracking data fast, easy to operate
The advantages of under, to the update classifier of different video frame adaptives, the die casting in different light environments can be adapted to
Defects detection, therefore applicability is wider.
The invention is not limited to above embodiment, if not departing from the present invention to various changes or deformation of the invention
Spirit and scope, if these changes and deformation belong within the scope of claim and equivalent technologies of the invention, then this hair
It is bright also to change and deform comprising these together.
Claims (10)
1. a kind of die casting defect inspection method, comprising:
Obtain the defect image data of die casting in actual production；
Extract the defects of defect image feature；
Classify to the defect characteristic；
Corresponding semantic label is established to the classification of the various defect characteristics；
The defect characteristic and corresponding semantic label are stored in property data base；
The video image on die casting surface is shot, and is converted into image data；
It closes filter tracks algorithm using nuclear phase to track the image data, with the defects of property data base spy
Sign is drawbacks of the standard feature, records the drawbacks of the standard feature having in the image data traced into；
Count ratio shared by the drawbacks of the standard characteristic and different drawbacks of the standard features of the image data.
2. a kind of die casting defect inspection method according to claim 1, it is characterised in that: in the property data base
Defect characteristic the step of being drawbacks of the standard feature include: that input needs the die casting defect title that detects, in property data base
The matching defect characteristic of middle search, using the defect characteristic searched as the drawbacks of the standard feature.
3. a kind of die casting defect inspection method according to claim 1, it is characterised in that: to the defect characteristic into
Before row classification, the defects of Wavelet Algorithm and image enhancement technique processing defect image feature are used；Wherein, the small echo
Denoising Algorithm includes carrying out denoising with defect image of the Nonlinear Wavelet Transform threshold method to die casting, with flexible and translation
Arithmetic operation carries out multiscale analysis to die casting image, automatic to adapt in different type die casting and light environment
It is required that showing the local feature of die casting；Described image enhancing technology includes enhancing algorithm with logarithmic image, distinguishes die casting figure
The defect and background of picture.
4. a kind of die casting defect inspection method according to claim 3, it is characterised in that: the nonlinear wavelet becomes
Changing the step of threshold method carries out denoising to the defect image of die casting includes: selection wavelet basis and the wavelet decomposition number of plies to pressure
The defect image of casting carries out wavelet transformation, obtains corresponding coefficient of wavelet decomposition, for each wavelet decomposition layer, adaptive
A threshold value is selected, obtained high frequency coefficient will be decomposed and carry out threshold value quantizing, according to the low frequency coefficient and threshold value after wavelet decomposition
High frequency coefficient after quantification treatment carries out wavelet inverse transformation, carries out wavelet reconstruction using restructing algorithm, obtains denoised signal.
5. a kind of die casting defect inspection method according to claim 1, it is characterised in that: carried out to the defect characteristic
The step of classifying, establishing corresponding semantic label to the classification of the various defect characteristics includes: to input the defect image
It is trained in Alexnet convolutional neural networks, the minimum value of loss function f (χ, b) is acquired using back-propagation algorithm, wherein
χ is weighted value, and b is biasing, using sigmoid functionAs activation primitive, declined by more gradients
Iterative method updates weighted value and acquires optimal weights value, inputs corresponding weighted value { χ according to different defect characteristics1,χ2,…
χn, shallow-layer extracts the preliminary feature of die casting defect, and deep layer extracts the minutia of die casting defect, training convolutional network mould
After type, classify to die casting defect characteristic, marks all kinds of die castings to lack with softmax classifier in the end of depth convolutional network
Feature is fallen into, corresponding semantic label is established.
6. a kind of die casting defect inspection method according to claim 1, it is characterised in that: described to utilize core correlation filtering
The step of device track algorithm tracks the image data includes: by following to die casting Bug Tracking target area
Ring displacement, constructs cluster sampling training classifier, calculates candidate region and tracks the similarity degree of target, it is maximum to choose similarity
Candidate region be new tracking target, utilize discrete Fourier transform to reduce the operation in classifier training and detection process
Amount.
7. a kind of die casting defect inspection method according to claim 6, it is characterised in that: the core correlation filter with
Track algorithm is to close filter tracks algorithm based on adaptive nuclear phase, and step includes: to be adjusted by being updated to classifier
Target is carried out piecemeal by the weighted value of whole present frame, calculates the PSR value of each piecemeal, each localized mass of adaptively changing is to mesh
The weight of cursor position.
8. a kind of die casting defect detecting device, comprising:
For obtaining the device of the defect image data of die casting in actual production；
For extracting the device of the defects of defect image feature；
Device for classifying to the defect characteristic；
The device of corresponding semantic label is established for the classification to the various defect characteristics；
For the defect characteristic and corresponding semantic label to be stored in the device of property data base；
For shooting the device of the video image on die casting surface and being translated into the device of image data；
For closing the device that filter tracks algorithm tracks the image data, and the institute that record traces by nuclear phase
State the device for the drawbacks of the standard feature having in image data；
For counting the dress of ratio shared by the drawbacks of the standard characteristic and different drawbacks of the standard features of the image data
It sets.
9. a kind of readable storage medium, stores control program thereon, it is characterised in that: when the control program is executed by processor
Realize die casting defect inspection method as claimed in any one of claims 1 to 7.
10. a kind of computer control system, including reservoir, processor and it is stored in the reservoir and can be by the place
Manage the control program that device executes, it is characterised in that: the processor realizes such as claim 1 to 7 when executing the control program
Die casting defect inspection method described in any one.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811056776.XA CN109509170B (en) | 2018-09-11 | 2018-09-11 | Die casting defect detection method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811056776.XA CN109509170B (en) | 2018-09-11 | 2018-09-11 | Die casting defect detection method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109509170A true CN109509170A (en) | 2019-03-22 |
| CN109509170B CN109509170B (en) | 2021-08-17 |
Family
ID=65745724
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811056776.XA Active CN109509170B (en) | 2018-09-11 | 2018-09-11 | Die casting defect detection method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109509170B (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110346953A (en) * | 2019-07-02 | 2019-10-18 | 盐城华昱光电技术有限公司 | A kind of the sheet detection system and method for the removing of liquid crystal display die set polaroid |
| CN110414538A (en) * | 2019-07-24 | 2019-11-05 | 京东方科技集团股份有限公司 | Defect classification method, defect classification based training method and device thereof |
| CN110726724A (en) * | 2019-10-22 | 2020-01-24 | 北京百度网讯科技有限公司 | Defect detection method, system and device |
| CN111161224A (en) * | 2019-12-17 | 2020-05-15 | 沈阳铸造研究所有限公司 | Casting internal defect grading evaluation system and method based on deep learning |
| CN111553542A (en) * | 2020-05-15 | 2020-08-18 | 无锡职业技术学院 | User coupon verification and sale rate prediction method |
| CN113267139A (en) * | 2021-07-19 | 2021-08-17 | 江苏中科云控智能工业装备有限公司 | Die casting deformation amount detection system with big data analysis |
| CN113284143A (en) * | 2021-07-20 | 2021-08-20 | 江苏中科云控智能工业装备有限公司 | Die casting deburring precision detection system based on image data processing |
| CN115456652A (en) * | 2022-09-29 | 2022-12-09 | 广东格林精密部件股份有限公司 | Tracing method for defective products of precision injection molding parts based on artificial intelligence |
| CN117392131A (en) * | 2023-12-12 | 2024-01-12 | 宁波昱辰汽车零部件有限公司 | Method and system for detecting defects of inner wall of die casting |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105118044A (en) * | 2015-06-16 | 2015-12-02 | 华南理工大学 | Method for automatically detecting defects of wheel-shaped cast product |
| CN105891215A (en) * | 2016-03-31 | 2016-08-24 | 浙江工业大学 | Welding visual detection method and device based on convolutional neural network |
| CN107123114A (en) * | 2017-04-21 | 2017-09-01 | 佛山市南海区广工大数控装备协同创新研究院 | A kind of cloth defect inspection method and device based on machine learning |
| CN108257114A (en) * | 2017-12-29 | 2018-07-06 | 天津市万贸科技有限公司 | A kind of transmission facility defect inspection method based on deep learning |
| WO2018154562A1 (en) * | 2017-02-23 | 2018-08-30 | D.I.R. Technologies (Detection Ir) Ltd. | A system and method for the inspection and detection of coating defects |
-
2018
- 2018-09-11 CN CN201811056776.XA patent/CN109509170B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105118044A (en) * | 2015-06-16 | 2015-12-02 | 华南理工大学 | Method for automatically detecting defects of wheel-shaped cast product |
| CN105891215A (en) * | 2016-03-31 | 2016-08-24 | 浙江工业大学 | Welding visual detection method and device based on convolutional neural network |
| WO2018154562A1 (en) * | 2017-02-23 | 2018-08-30 | D.I.R. Technologies (Detection Ir) Ltd. | A system and method for the inspection and detection of coating defects |
| CN107123114A (en) * | 2017-04-21 | 2017-09-01 | 佛山市南海区广工大数控装备协同创新研究院 | A kind of cloth defect inspection method and device based on machine learning |
| CN108257114A (en) * | 2017-12-29 | 2018-07-06 | 天津市万贸科技有限公司 | A kind of transmission facility defect inspection method based on deep learning |
Non-Patent Citations (2)
| Title |
|---|
| MAX FERGUSON等: ""Automatic localization of casting defects with convolutional neural networks"", 《2017 IEEE INTERNATIONAL CONFERENCE ON BIG DATA (BIG DATA)》 * |
| 郑晓玲等: ""采用机器视觉的铝压铸件表面缺陷检测"", 《华侨大学学报（自然科学版）》 * |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110346953A (en) * | 2019-07-02 | 2019-10-18 | 盐城华昱光电技术有限公司 | A kind of the sheet detection system and method for the removing of liquid crystal display die set polaroid |
| CN110346953B (en) * | 2019-07-02 | 2022-08-09 | 晋城市龙鑫达光电科技有限公司 | Tearing detection system and method for stripping polaroid of liquid crystal display module |
| US11334982B2 (en) | 2019-07-24 | 2022-05-17 | Beijing Boe Optoelectronics Technology Co., Ltd. | Method for defect classification, method for training defect classifier, and apparatus thereof |
| CN110414538A (en) * | 2019-07-24 | 2019-11-05 | 京东方科技集团股份有限公司 | Defect classification method, defect classification based training method and device thereof |
| CN110414538B (en) * | 2019-07-24 | 2022-05-27 | 京东方科技集团股份有限公司 | Defect classification method, defect classification training method and device thereof |
| CN110726724A (en) * | 2019-10-22 | 2020-01-24 | 北京百度网讯科技有限公司 | Defect detection method, system and device |
| CN111161224A (en) * | 2019-12-17 | 2020-05-15 | 沈阳铸造研究所有限公司 | Casting internal defect grading evaluation system and method based on deep learning |
| CN111553542A (en) * | 2020-05-15 | 2020-08-18 | 无锡职业技术学院 | User coupon verification and sale rate prediction method |
| CN111553542B (en) * | 2020-05-15 | 2023-09-05 | 无锡职业技术学院 | User coupon verification rate prediction method |
| CN113267139A (en) * | 2021-07-19 | 2021-08-17 | 江苏中科云控智能工业装备有限公司 | Die casting deformation amount detection system with big data analysis |
| CN113284143A (en) * | 2021-07-20 | 2021-08-20 | 江苏中科云控智能工业装备有限公司 | Die casting deburring precision detection system based on image data processing |
| CN115456652A (en) * | 2022-09-29 | 2022-12-09 | 广东格林精密部件股份有限公司 | Tracing method for defective products of precision injection molding parts based on artificial intelligence |
| CN117392131A (en) * | 2023-12-12 | 2024-01-12 | 宁波昱辰汽车零部件有限公司 | Method and system for detecting defects of inner wall of die casting |
| CN117392131B (en) * | 2023-12-12 | 2024-02-06 | 宁波昱辰汽车零部件有限公司 | Method and system for detecting defects of inner wall of die casting |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109509170B (en) | 2021-08-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109509170A (en) | A kind of die casting defect inspection method and device | |
| CN108765412B (en) | Strip steel surface defect classification method | |
| CN107563999A (en) | A kind of chip defect recognition methods based on convolutional neural networks | |
| CN110533100A (en) | A method of CME detection and tracking is carried out based on machine learning | |
| CN106340007A (en) | Image processing-based automobile body paint film defect detection and identification method | |
| CN109711326A (en) | A kind of video object detection method based on shallow-layer residual error network | |
| Hussein et al. | Reconstruction of damaged herbarium leaves using deep learning techniques for improving classification accuracy | |
| CN116612335A (en) | Few-sample fine-granularity image classification method based on contrast learning | |
| CN108154513A (en) | Cell based on two photon imaging data detects automatically and dividing method | |
| CN106650824B (en) | Moving object classification method based on support vector machines | |
| CN108288283A (en) | A kind of video tracing method based on correlation filtering | |
| Yu et al. | Research on CNN Algorithm for Monochromatic Fabric Defect Detection | |
| CN117372332A (en) | Fabric flaw detection method based on improved YOLOv7 model | |
| CN116735610A (en) | Steel pipe surface defect detection method based on machine vision | |
| CN109376619A (en) | A kind of cell detection method | |
| CN109117818A (en) | Material structure characteristic intelligent recognition analysis system and analysis method | |
| CN113962968A (en) | Multi-source mixed interference radar image target detection system oriented to complex electromagnetic environment | |
| Jadah et al. | Breast Cancer Image Classification Using Deep Convolutional Neural Networks | |
| CN104573746A (en) | Fruit fly type identification method based on magnetic resonance imaging | |
| Krithiga et al. | Proliferation score prediction using novel SMHC feature using adaptive XGBoost model | |
| CN118277674B (en) | Personalized image content recommendation method based on big data analysis | |
| CN118097141B (en) | Obstetrical image-based image segmentation method and system | |
| Cui et al. | Automatic Identification Technology of Optical Fiber based on Genetic Neural Network Algorithm | |
| Nimitha et al. | Diagnosis and treatment recommendation engine for skin diseases using AI model VGG19 | |
| Zhang et al. | Defect detection of color-patterned fabric based on DenoisingGAN |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |